Quantifying error with pixel ilastik classifier?

Hi all,

I’m using a pixel classifier in iLastik to distinguish a structure from the background. I then use the number of pixels in that structure to measure its volume. My question is: how can I estimate the error that the segmentation induces?

Thank you!
Anjalie Schlaeppi

Hi Anjalie,

Have you already created a set of ground truth images?

Ed

Dear Ed,

I don’t but I could manually segment a few images points and compare the number of pixels enclosed.

Each of my data sets have around 20’000 images, do you suggest talking a few representative of the whole stack to manually segment?

Thank you!
Anjalie

Hi Anjalie,

Yes I think creating some ground truth manually would be the only way to solve your problem at the moment. There are methods in CellProfiler software to automate the counting of pixels in your ground truth vs ilastik segmented images. Have you tried that?

Ed

1 Like

This may be of help:

1 Like

Awesome, this is super helpful, thank you so much!

Cheers,
Anjlie